The plot_directive Sphinx extension imports both NumPy and Matplotlib, by default. Great for inline code examples, maybe not so great for auto-generated-from-standalone examples.

I suspected something like this… I guess it is a really good thing no one else than matplotlib uses the plot_directive.

If no one uses it perhaps we should just not add that header anymore :-)

If someone fixes that file, two more easy fixes:

• I think the meddling with _dummy_print is probably not useful as we already redirect sys.stdout.
• the try: try: ... except: ... finally: ... can be flattened into a single try: ... except: ... finally: ... (since py2.5... pep341)

I use the plot directive quite a bit in my day-job docs and it is in both the numpy and scipy docs.

Suspect the fix here is to make sure this file is run by the backend-driver test.

I would rather deprecate this.

You can do something like

• try to run without the imports
• if it fails, try to run with the imports and raise a deprecationwarning if that works.

I am 👎 on deprecating this. The plot directive is very useful for in-line plots where reducing boiler plate is generally good and changing this is going to cause a bunch of work for down-stream users.

If we were writing it now no default imports would be reasonable, but it has been this way from 2010 (6f2860f) and I don't see the payoff being worth the down-stream cost.

If we want to be strict in our docs, add a 'strict' flag that disables the imports (or just pass an empty string into plot_pre_code in the templates we use to generate the gallery).

plot_pre_code is exactly there to avoid the need for boilerplate.

If we finish the migration to sphinx-gallery, I don't think this function will be necessary anymore.
Anyhow, we know why we have some examples that run inside our galleries and outside, so I suggest closing this ticket.

@NelleV can you close this? I think that this issue is resolved since sphinx-gallery is building all of the examples etc.